https://schristoph.online/tags/inferenceoptimization/Personal homepage and blog of Stefan ChristophHugo -- gohugo.ioen-usStefan Christoph. All rights reserved.Sat, 21 Mar 2026 00:00:00 +0000https://schristoph.online/blog/when-thinking-twice-helps/?utm=rss-feedSat, 21 Mar 2026 00:00:00 +0000https://schristoph.online/blog/when-thinking-twice-helps/<h2 id="the-saturday-morning-experiment">The Saturday Morning Experiment</h2> <p>Last Saturday, I installed a Python library, pointed it at Amazon Bedrock, and asked a model the same questions three times, with zero, one, and three rounds of self-reflection.</p> <p>The results surprised me.</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-fallback" data-lang="fallback"><span class="line"><span class="cl"> Q Refl Time Acc Comp Nuan Total </span></span><span class="line"><span class="cl"> 1 0 3.0s 4 3 3 10 </span></span><span class="line"><span class="cl"> 1 1 5.5s 4 2 3 9 </span></span><span class="line"><span class="cl"> 1 3 8.8s 4 3 4 11 </span></span><span class="line"><span class="cl"> 2 0 2.6s 4 2 2 8 </span></span><span class="line"><span class="cl"> 2 1 5.7s 4 2 2 8 </span></span><span class="line"><span class="cl"> 2 3 8.5s 4 2 2 8 </span></span><span class="line"><span class="cl"> 3 0 3.1s 1 1 1 3 </span></span><span class="line"><span class="cl"> 3 1 5.2s 1 1 1 3 </span></span><span class="line"><span class="cl"> 3 3 8.6s 1 1 1 3 </span></span></code></pre></div><p>Q is the question number, Refl the number of self-reflection rounds (0 = straight answer, 1 = one revision, 3 = three revisions). Acc, Comp, and Nuan are the judge&rsquo;s scores for Accuracy, Completeness, and Nuance, each on a 1-5 scale, 15 max total.</p>